FURI | Fall 2020

Evaluating Collective Intelligence Strategies for Visual Detection

This work investigates how different methods of soliciting and aggregating input from multiple people best inform visual detection tasks. A crowdsourced experiment is developed which asks hundreds of study participants are asked to identify if an object is or is not in an image. Then, aggregation techniques such as majority voting, confidence weighted voting, and optimization-based aggregation models are tested to see how well they perform at determining correct classifications from multiple noisy answers. The results of this project seek to inform best-practice methodologies of using collective intelligence to inform object detection.